Re: problem with the "write_bw_log" parameter in a virtualized environment using a write iopattern

[Jens Axboe <jens.axboe@oracle.com>]

On Fri, Jul 31 2009, Ivan Frain wrote:
> Jens Axboe a �crit :
>> On Fri, Jul 31 2009, Ivan Frain wrote:
>>   
>>> Hi everybody,
>>>
>>> I'am using fio in a vmware virtualized environment. I have got a guest
>>> virtual machine based on centos5.2 on which I run fio in order to
>>> measure the virtual disks IO performances.
>>> I use the "write_bw_log" parameter in order to build graphs for the
>>> experiments but the problem is that the file generated by fio for each
>>> job is empty (size = 0).
>>> However, i also use "write_clat_log" parameter and the 'clat.log' and
>>> 'slat.log' files are not empty. The mean bandwidth in the fio output
>>> summary seams to be good.
>>>     
>>
>> I wonder how long that bug has been there... It's due to a botched
>> memcpy in the fio core. There are two stat sample times, yet it
>> overwrites only the first one (being 0, or DDIR_READ, which explains why
>> the read one works).
>>
>> Try the below patch, it's also committed now.
>>
>> diff --git a/fio.c b/fio.c
>> index ba9e384..7ad1b57 100644
>> --- a/fio.c
>> +++ b/fio.c
>> @@ -1105,7 +1105,10 @@ static void *thread_main(void *data)
>>  	clear_state = 0;
>>  	while (keep_running(td)) {
>>  		fio_gettime(&td->start, NULL);
>> -		memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start));
>> +		memcpy(&td->ts.stat_sample_time[0], &td->start,
>> +				sizeof(td->start));
>> +		memcpy(&td->ts.stat_sample_time[1], &td->start,
>> +				sizeof(td->start));
>>  		memcpy(&td->tv_cache, &td->start, sizeof(td->start));
>>   		if (td->o.ratemin[0] || td->o.ratemin[1])
>>
>>   
>
> Well, this patch works very well, the bw log file is filled again.
> Thank you for your quick answer.

Good!

> FIO rocks !!

Thanks :-)

-- 
Jens Axboe